ARTICLE

Application of multistate modeling to estimate salmonidsurvival and movement in relation to spatial and temporalvariation in metal exposure in a mining-impacted riverMariah P. Mayfield, Thomas E. McMahon, Jay J. Rotella, Robert E. Gresswell, Trevor Selch, Patrick Saffel,Jason Lindstrom, and Bradley Liermann

Abstract: Multistate modeling was used to estimate survival and movement of brown trout (Salmo trutta) and westslope cutthroattrout (Oncorhynchus clarkii lewisi) in relation to copper concentrations in the mining-impacted Clark Fork River, Montana. Survivalprobability in the uppermost river segment, where dissolved copper concentrations frequently exceeded acute criteria foraquatic life (range: 31–60 days > 13.4 g·L–1), was 2.1 times lower for brown trout and 122 times lower for westslope cutthroat troutcompared with survival rates in the lowermost segment that had relatively low dissolved copper (0 days exceedance of acuteconcentration). Lowest survival for both species occurred in the spring–summer period when dissolved copper concentrationswere elevated coincident with higher discharge. Movement among study segments was generally low, and cutthroat troutshowed low movement into the uppermost river segment with the most elevated copper levels. Both species showed high ratesof movement into tributaries, which coincided with their respective spawning migrations rather than as an apparent avoidanceof elevated copper levels. The linkage between survival rate and level of copper exposure for both trout species suggests thatadditional removal of tailings deposits could improve survival rates.

Résumé : La modélisation multi-états est utilisée pour estimer la survie et les déplacements de truites brunes (Salmo trutta) et detruites fardées versant de l’ouest (Oncorhynchus clarkii lewisi) au vu des concentrations de cuivre dans la rivière Clark Fork(Montana), une rivière touchée par les impacts de l’activité minière. La probabilité de survie dans le tronçon le plus amont de larivière, où les concentrations de cuivre dissous dépassent fréquemment les critères d’exposition aigüe pour la vie aquatique(fourchette: 31–60 jours > 13,4 g·L–1), était 2,1 fois plus faible pour la truite brune et 122 fois plus faible pour la truite fardéeversant de l’ouest que les taux de survie dans le tronçon le plus aval, caractérisé par des concentrations de cuivre dissousrelativement faibles (0 jour de dépassement de la concentration associée à une exposition aigüe). Les taux de survie les plusfaibles pour les deux espèces ont été observés durant la période printanière et estivale, moment où de fortes concentrations decuivre dissous coïncidaient avec des débits plus importants. Il y avait généralement peu de déplacements entre les tronçonsétudiés, et les truites fardées versant de l’ouest présentaient des déplacements limités dans le tronçon le plus amont de la rivièrecaractérisé par les plus fortes concentrations de cuivre. Les deux espèces présentaient des fréquences élevées de déplacementsvers les affluents, qui coïncidaient avec leurs migrations de frai respectives plutôt que de refléter d’apparents efforts pour éviterde fortes concentrations de cuivre. Le lien entre le taux de survie et le niveau d’exposition au cuivre pour les deux espèces detruite donne à penser que la poursuite du retrait des dépôts de stériles pourrait améliorer les taux de survie. [Traduit par laRédaction]

IntroductionWater pollution from mining is one of the most detrimental

and persistent anthropogenic impacts in fresh waters. Mining canrelease of large volumes of trace metals and waste material,which, in turn, can adversely affect water quality, physical habi-tat, fish populations, and aquatic food webs over extensive areasfor long time periods (Farag et al. 2003; Luoma et al. 2008; Kiseret al. 2010). The US Environmental Protection Agency (EPA) rank-ing of hazardous waste sites designates the most contaminatedsites as “Superfund” sites, so named for their particularly harmfuland far-reaching environmental impacts (Moore and Luoma 1990).One such example is the upper Clark Fork River in southwestern

Montana, the largest Superfund site in the United States, whereoperation of the Butte copper mines from the 1880s to 1960s re-sulted in deposition of an estimated 99.8 billion kg of miningwaste in the floodplain, rendering about 200 km of the river es-sentially devoid of fish for many decades due to metal toxicity(Phillips and Lipton 1995).

Elevated metal concentrations affect aquatic ecosystems atmultiple levels of biological organization (Clements 2000). At theindividual level, metal exposure can contaminate prey and bioac-cumulate in fish tissues, resulting in a multitude of physiologicalresponses that diminish health and reduce survival, includingimpaired feeding, food assimilation, and growth (Woodward et al.

Received 18 July 2018. Accepted 10 January 2019.

M.P. Mayfield,* T.E. McMahon, and J.J. Rotella. Ecology Department, Fish and Wildlife Ecology and Management Program, Montana State University,Bozeman, MT 59717, USA.R.E. Gresswell. US Geological Survey, Northern Rocky Mountain Science Center, 2327 University Way, Bozeman, MT 59715, USA.T. Selch. Montana Fish, Wildlife and Parks, 1420 E. Sixth Avenue, Helena, MT 59620, USA.P. Saffel, J. Lindstrom, and B. Liermann. Montana Fish, Wildlife and Parks, 3201 Spurgin Road, Missoula, MT 59804, USA.Corresponding author: Thomas E. McMahon (email: tmcmahon@montana.edu).*Present address: USDA Forest Service, Okanogan-Wenatchee National Forest, 24 West Chewuch Road, Winthrop, WA 98862, USA.Copyright remains with the author(s) or their institution(s). Permission for reuse (free in most cases) can be obtained from RightsLink.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

1

Can. J. Fish. Aquat. Sci. 00: 1–12 (0000) dx.doi.org/10.1139/cjfas-2018-0280 Published at www.nrcresearchpress.com/cjfas on 7 February 2019.

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

1995a; Farag et al. 2000). Direct mortality from metal exposure canoccur when metals are mobilized from stream bank and flood-plain deposits during high flow events and exceed acute toxicitylevels (Marr et al. 1995; Phillips and Lipton 1995; Luoma et al.2008). Furthermore, behavioral avoidance of waters with elevatedconcentrations of metals can limit distribution and access to suit-able habitat and may occur at concentrations substantially lowerthan those causing acute effects on survival (Woodward et al.1995a; Goldstein et al. 1999; Svecevicius 2001). At higher levels oforganization, changes in fish assemblages occur in response toelevated metal concentrations when species have different phys-iological sensitivities and behavioral avoidance thresholds to met-als. For example, in the western United States, nonnative browntrout (Salmo trutta) are less sensitive to metals than cutthroat trout(Oncorhynchus clarkii) and rainbow trout (Oncorhynchus mykiss), andthis variable sensitivity has resulted in a shift toward brown troutdominance in mining-contaminated systems (Hansen et al. 1999a).

Prior to mining, native bull trout (Salvelinus confluentus) and west-slope cutthroat trout (O. clarkii lewisi) were widespread throughoutthe upper Clark Fork River drainage (Luoma et al. 2008). However,surveys in the late 1800s to the 1950s found very few or no fish inthe river from Silver Bow Creek, near Butte, Montana, down-stream to Rock Creek, near Missoula, Montana (Phillips andLipton 1995). Water, sediments, and aquatic invertebrates con-tained high levels of copper, zinc, lead, cadmium, and arsenic(Axtmann and Luoma 1991). A series of remediation efforts beganin the 1960s to alleviate the primary source of trace metals, SilverBow Creek, a headwater tributary of the upper Clark Fork River(Fig. 1; Phillips and Lipton 1995). These activities included instal-lation of liming ponds and a wastewater treatment plant and,more recently, removal of stream channel and floodplain sedi-ments containing high metal concentrations. Following remedia-tion in Silver Bow Creek, metal concentrations downstream in theupper Clark Fork River declined substantially (Luoma et al. 2008;Hornberger et al. 2009), and trout began to recolonize in the 1970safter an ostensibly century-long absence (Phillips and Lipton1995). Nonnative brown trout are now the dominant trout species,

but nonnative rainbow trout and native westslope cutthroat troutand bull trout, which are low to moderately abundant in nearby,uncontaminated rivers, are rare (Luoma et al. 2008). Currently,metal concentrations in sediments in the upper Clark Fork Riverdecline exponentially downstream (Axtmann and Luoma 1991;Luoma et al. 2008; Hornberger et al. 2009), as do metal concentra-tions in fish and macroinvertebrate tissues (Farag et al. 1995).However, dissolved metal concentrations, especially copper, con-tinue to periodically exceed acute criteria for aquatic life (13.4 g·L−1;USEPA 2007) during pulses of high discharge during spring runoffand summer thunderstorms (Nimick and Moore 1991; Luoma et al.2008).

Trout population estimates conducted in the 1990s (summa-rized in Luoma et al. 2008) suggested mean trout densities in theupper Clark Fork River were about 10% of expected trout densitycompared with noncontaminated reference reaches in nearby riv-ers. More recent surveys, using similar sampling methodology inthe same initial sampling reaches, suggested similar trout densi-ties (Lindstrom 2011; Cook et al. 2015). Thus, although there hasbeen widespread rebound of trout populations in the upper ClarkFork River after remediation, continued periodic pulses of ele-vated copper concentrations exceeding the acute criteria leveland incomplete recovery of trout population density to premininglevels suggest that continued mortality might still be inhibitingtrout population recovery, particularly for more metal-sensitivespecies like cutthroat trout. Additional limiting habitat factorspresent in the river basin (e.g., elevated temperature and move-ment barriers) might also contribute to reduced trout abundance(Clark Fork Coalition 2011; Cook et al. 2015).

We examined factors limiting trout abundance in the upperClark Fork River by estimating survival and movement rates ofbrown trout and westslope cutthroat trout using telemetry data ina multistate model. Multistate modeling facilitates estimation ofsurvival and movement relative to different “states” (Buchananand Skalski 2010; Perry et al. 2010); in our study, we examined howtrout survival and movement rates varied in relation to spatial andtemporal differences in copper exposure. We assessed whether trout

Fig. 1. Three mainstem segments, major tributaries and towns, and discharge, temperature, and metal sampling gauge stations, upper ClarkFork River, Montana, 2009–2011.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

2 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

in the upper Clark Fork River exhibit lower survival and higheravoidance rates (i) in areas with higher copper concentrations, spe-cifically the upper 63 km of the river where mining tailing depos-its are still widespread compared with downstream sections withlower exposure and (ii) during time periods of increased copperexposure that likely occur during spring and summer when snow-melt or rainfall events increase stream discharge and mobilizecopper into the water column from sediments (Luoma et al. 2008).Additionally, we assessed whether survival and movement re-sponses to copper exposure are more pronounced in cutthroattrout due to their heightened metal sensitivity (Woodward et al.1995b).

Methods

Study areaThe study area included the upper Clark Fork River, from the

confluence of Silver Bow and Warm Springs creeks, near WarmSprings, Montana, downstream to the confluence with the Black-foot River, near Missoula, Montana (189 river kilometres), andassociated tributaries (Fig. 1). As noted, extensive metal contami-nation from upstream copper mining that occurred from the1880s to the 1960s had major negative effects on the upper ClarkFork River, but tributaries exhibited little or no metal contamina-tion (Luoma et al. 2008). We divided the mainstem river studysection into three segments (Fig. 1) based on segment boundariesused in previous studies that assessed spatial patterns in sedimentmetal concentrations (Axtmann and Luoma 1991; Hornbergeret al. 2009): Segment A was a 63 km-long section from near theheadwaters of the Clark Fork River at Warm Springs downstreamto the Little Blackfoot River confluence; Segment B (42 km) wasbounded downstream by the Flint Creek confluence; and SegmentC (84 km) extended downstream to the Blackfoot River conflu-ence. Downstream segment boundaries were junctions of majortributaries that dilute copper concentrations in the main stem(Axtmann et al. 1997; Luoma et al. 2008).

Physical habitat and water qualityAmong the three study segments, we compared physical habi-

tat, temperature, and copper exposure. For physical habitat, wemeasured floodplain width and area every kilometre based on the50-year flood boundary using ArcMap Version 9.3.1 (Esri 2009). Thearea of streamside tailing deposits for each segment (Brooks and

Moore 1989) was summed from existing GIS data (Montana StateLibrary, GIS Layers for Clark Fork Superfund Site; https://mslservices.mt.gov/Geographic_Information/Data/DataList/datalist_Details.aspx?did=33b14155-3366-42a0-b50c-1e21d5224fe0) and divided by totalfloodplain area to obtain percentage of floodplain contaminationper segment. Dissolved copper concentration was determinedfrom metal data in water samples collected eight times per year atUS Geological Survey (USGS) discharge gauge stations locatedwithin each study segment (Fig. 1). The relation between copperconcentration and discharge was calculated for each gauge sta-tion across all 3 sampling years (n = 23–39) separately using simplelinear regression (see Hornberger et al. 2009). The resulting regres-sions (Fig. 2) were then used to predict daily copper concentra-tions in each segment based on daily discharge (Fig. 3). Annualacute copper exposure in each segment was calculated as thenumber of days exceeding the EPA acute dissolved copper concen-tration to protect aquatic life (13.4 g·L−1; USEPA 2007), based onthe observed water hardness of 100 mg·L−1 in the upper Clark ForkRiver. Mean daily copper exposures were also calculated by season(exact dates listed below) and compared among segments. In trib-utaries, dissolved copper concentration was derived from USGSwater samples collected intermittently from 2000 to 2011 fromthree relatively uncontaminated reference tributaries: LittleBlackfoot River, Flint Creek, and Rock Creek. Maximum daily tem-perature for each segment was determined from thermographdata collected at each of the USGS gauge stations.

RadiotelemetrySurvival and movements of brown trout and westslope cutthroat

trout were assessed using radio transmitters that were implantedinto trout systematically over the study area at a distribution of1 tag per 0.8 river kilometres. A total of 183 adult brown troutand 70 westslope cutthroat trout (282–570 mm total length) wereradiotagged. Cutthroat trout were identified as pure westslopecutthroat trout or slightly hybridized with rainbow trout based onvisual characteristics (Weigel et al. 2002). Fish were captured by aboat-mounted electrofisher in April, prior to spring runoff, in2009 (brown trout, n = 70; cutthroat trout, n = 21), 2010 (browntrout, n = 59; cutthroat trout, n = 33), and 2011 (brown trout, n = 9;cutthroat trout, n = 11). An additional 45 brown trout and fivecutthroat trout were radiotagged in September 2010 in the mainstem and in the lower reaches of Flint Creek and the Little Black-

Fig. 2. Relationships between dissolved copper concentration and discharge (Q) for the three mainstem segments on the upper Clark ForkRiver, Montana (2009–2011 combined). The dashed line is the acute toxicity standard for copper by the EPA (13.4 g·L–1 at water hardness of100 mg·L−1; USEPA 2007). Caution should be used in extrapolating dissolved copper concentrations to discharges beyond minimum andmaximum observations.

Discharge (m3·s-1)

0 100 200 300

(noitartnecno

C re ppoC

de vlo ssiD

µL

g-1

)

0

10

20

30

40

50

Segment A, y = -0.84 + 0.76(discharge), r2 = 0.88Segment B, y = 2.45 + 0.15(discharge), r2 = 0.79Segment C, y = 2.37 + 0.03(discharge), r2 = 0.79EPA Acute Copper Standard

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Mayfield et al. 3

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

foot River to assess survival and movement in tributaries andpotential seasonal differences in post-tagging survival.

Trout were anesthetized with tricaine methanesulfonate, mea-sured to the nearest millimetre, weighed to nearest 0.1 g, andimplanted with 8 or 10 g internal transmitters (dependent on bodyweight) with external antennas (Lotek Wireless, Newmarket, On-tario, 148.500 to 148.560 MHz). Tags were inserted in small inci-sions made dorsally to the pelvic girdle. The trailing externalantenna was extruded through a small hole made dorsally to thevent via a shielded hollow needle (Ross and Kleiner 1982). Incisionswere closed with sutures and instruments sterilized between proce-dures to minimize risk of infection. The mean relative weight(Anderson and Neumann 1996) of tagged fish was 93 (range: 66–121)for brown trout and 101 for cutthroat trout (range: 45–138). Tagweight was less than the recommended 4% of total body weight(Zale et al. 2005). Radiotags were programmed to a 12 h on : 12 h offschedule to prolong battery life (685 to 877 days). Following sur-gery, fish were placed in a live cage until they regained equilib-rium (10 min) and were released near the point of capture.

Radiotag relocations were obtained weekly over the entirestudy area from April 2009 to December 2011 (138 weeks), exceptduring winter months (January and February) when surveys wereconducted every 2 weeks. Relocations were made primarily fromroads using a combination of a vehicle-mounted omnidirectionalantenna and a directional three-element Yagi antenna. Areas withlimited road access were surveyed by boat. Trout position wastriangulated using a Lotek SRX 400 receiver and coordinates re-corded with a global positioning unit. Relocation accuracy wasestimated to be 100 m based on trials using dummy tags. Reloca-tion error was considered minimal given that our focus was onmovement among long study segments (42 to 84 km long). Fishlocation was converted to distance upstream from the lowerboundary of the study area (Blackfoot River confluence) usingArcMap.

Sixty-seven percent (171 of 256) of radiotags were equipped withmotion (mortality) sensors; mortality was assumed to occur whenunique radio signals emitted by motion sensors (indicating nomovement) were detected. For trout without mortality-sensor ra-diotags, we assumed that mortality had occurred if a given tag wasrelocated in the same location for several weeks and no move-ment of the tag was detected when investigators prodded andwaded through the tag relocation site.

Survival and movement analysisWe first plotted Kaplan–Meier survival curves (Kaplan and

Meier 1958) for each species and each of the 3 study years over the

entire study area. This analysis was used to examine overall sea-sonal patterns of survival associated with discharge and to assessevidence of mortality that might have occurred shortly after tag-ging and thereby biased our evaluation of the relationship be-tween survival and metal contamination levels.

Survival and movement were then examined in more detailusing multistate mark–recapture analysis of telemetry data(Arnason 1972, 1973; Hestbeck et al. 1991; Brownie et al. 1993)following the approach of Buchanan and Skalski (2010) and Perryet al. (2010). During each (typically weekly) tag–relocation period,a relocated trout was categorized as being in one of five “states”:the first four states were assigned to live fish and indicatedwhether they were occupying one of the three mainstem studysegments (A, B, or C) or were in a tributary (T). The fifth state wasassigned to fish that died during the relocation period anywherein the study area (“dead state”; D). A separate tributary state wasdesignated because tributaries had distinctly different habitatcharacteristics than mainstem habitat segments (much lower cop-per concentrations and the presence of irrigation structures thatcould restrict movement or affect survival). Angling pressure andangling-related mortality are low in the study area and thus wereunlikely to influence survival rates (P. Saffel, J. Lindstrom, and B.Liermann, unpublished data).

Weekly encounter histories were created for each individualbased on the state determined for the week. Fish that were lastencountered alive and that went undetected in a given week wereassigned a 0 for that week. For the small number of cases wherewe had evidence that transmitter batteries had likely failed priorto the end of the study, we truncated the encounter histories afterthe failure occurred. The multistate model provided estimates ofthe weekly probabilities of transitioning from one state to anyother state. The weekly transition probabilities, which combinedinformation on survival and movement (Buchanan and Skalski2010), indicated whether a fish (i) remained alive in its currentlocation (river segment or tributary); (ii) remained alive andmoved to another location; or (iii) died. The model allowed thetransition probabilities to vary by the location occupied at the start ofthe week, by species, and over different time periods. The weeklyprobability of remaining alive was the complement of transition-ing to the dead state. For example, for an individual starting theweek in Segment A, the weekly probability of mortality was theprobability of transitioning from states A to D; the weekly proba-bility of staying alive somewhere in the study area was the sum oftransition probabilities A to A, A to B, A to C, and A to T. We thenused weekly survival estimates to compare seasonal and annual

Fig. 3. Predicted dissolved copper concentrations (g·L−1) in relation to discharge in the upper Clark Fork River, Montana, for study SegmentsA (Deer Lodge sampling station), B (Gold Creek), and C (Turah), using linear regression equations from Fig. 2. The dashed line is the acutetoxicity standard for dissolved copper by the EPA (13.4 g·L−1) at water hardness of 100 mg·L−1 (USEPA 2007).

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

4 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

survival rates for brown trout and westslope cutthroat trout occu-pying each study segment (A, B, C, or T). Seasonal survival esti-mates were calculated as the product of the weekly survivalestimate for the season in question. Annual survival estimateswere calculated as the product of the seasonal survival estimates.The standard errors (SEs) for seasonal and annual survival esti-mates were estimated using the delta method (Powell 2007) inProgram R (R Development Core Team 2010) and the “msm” pack-age (Jackson 2011).

A set of competing multistate models that constrained transi-tion probabilities through time and space were developed to fur-ther assess evidence for differences in survival and movementrates among the four different spatial states (three mainstem seg-ments and all tributaries) and among four different time states(seasons). Seasonal differences in survival and movement weredetermined by grouping weekly transition rates into four bio-logically meaningful time periods: spring (20 March to 15 June),characterized by high discharge and elevated dissolved copperconcentrations (>10 g·L−1); summer (16 June to 15 September),characterized by low discharge, low copper concentrations (3–6 g·L−1), and higher temperatures interspersed with occasionalpulses of high copper concentrations during elevated dischargesfrom summer thunderstorms (Luoma et al. 2008); fall (16 Septem-ber to 20 December), characterized by low discharge, declining tem-perature, and low copper concentrations; and winter (21 Decemberto 20 March), characterized by low discharge, temperature, and cop-per and presence of ice cover. Additionally, we also tested a single-season “spring only” model, which allowed spring transition ratesto vary while other seasons were held constant, and a “combinedseason” model that combined spring and summer transition ratesand fall and winter transition rates. The spring only model wasdeveloped to examine if the higher copper concentrations duringspring runoff (>10 g·L−1) were the most influential on survivaland movement patterns. The combined season model was devel-oped to examine if the presumed elevated copper levels in springand summer and low levels in fall and winter better explainedobserved survival and movement than a full four-season model. Inall, ten different combinations of survival and movement ratemodels were evaluated, with independent estimates computedfor brown trout and cutthroat trout separately (Appendix A,Table A1). For all models, we combined data from all 3 years of thestudy because temporal trends were similar among years.

Models were run in Program MARK (White and Burnham 1999).Prior to model selection, we evaluated goodness of fit for our mostcomplex model using the median c procedure in Program MARK(White and Burnham 1999). Because there was no evidence foroverdispersion among any of the models tested (median c = 0.99,95% CI: 0.97–1.01), we based model selection on AICc (adjusted forsample size) rather than QAICc values (Burnham and Anderson2002). We considered the model with the lowest value to be thebest and evaluated the plausibility of other models based on thedifference between the AICc value for the top model and that of everyother model (AICc). For a model with AICc values of <2, 4–7,and >10, there is substantial, some, and no evidence, respectively,that differences were statistically significant (Burnham andAnderson 2002).

We subsequently used the best approximating model (com-bined season) to test whether degree of copper exposure couldpotentially influence movement and survival. To do this, we inputestimates of weekly survival and movement into a matrix popu-lation projection model (Caswell 2001) to calculate (i) the numberof weeks that a fish would remain alive from the start of the yearconditional on its starting location (A, B, C, or T) and (ii) how manyweeks it would spend in each segment while it remained alive. Asin our multistate model, the possible transitions that could be madeeach week consisted of a fish (1) remaining alive in its current loca-tion (river segment or tributary); (2) remaining alive and moving toanother location; or (3) dying. The probabilities of the weekly tran-

sitions were determined state of an individual at the start of theweek. If an individual moved to a new location, the transitionprobabilities estimated by the multistate model for that locationwere used to project that individual in the next week.

The weekly transition probabilities were also allowed to vary byweek in accordance with the temporal structure of weekly survivalrates from the combined season model. For example, differenttransition rates were used in projection model for the 26-week-long spring–summer and fall–winter periods.

Finally, to incorporate the uncertainty that was associated withestimating the transition probabilities, we used Monte Carlo simula-tions of the projection model, where each simulation consisted ofprojecting the location and live or dead status of one trout forwardfor 10 years. We simulated projections for 400 000 trout, with100 000 starting out in each river segment. The actual transitionprobabilities used in each step of each projection for a given weekand location were randomly drawn based on the estimated meanand variance from the combined season model; the log-odds val-ues were then back-transformed to the 0–1 probability scale foruse in the simulation. All calculations were completed using the“popbio” package in Program R (Stubben and Milligan 2007).

Results

Habitat and water qualityStudy segments differed substantially in extent of mining tail-

ings deposits and dissolved copper concentration. Segment A hada wide floodplain (mean ± SE: 704 ± 43 m) with extensive miningtailing deposits (63.3 ha; mean 1 ha tailings·river km−1; 2.3% oftotal floodplain area) and dissolved copper concentrations thatregularly exceeded EPA acute standards for dissolved copper(13.4 g·L−1; USEPA 2007), especially during high discharge, withpeak copper levels of 24–46 g·L−1 during the 3 study years (Fig. 3).Segment B also had a wide floodplain (686 ± 115 m) but containedmuch less extensive floodplain tailings (15.6 ha; mean 0.4 ha·km−1;0.1% of floodplain) and peak dissolved copper concentrations dur-ing the 3 study years (14–29 g·L−1) that exceeded acute levels onlyduring peak discharge events. Segment C had a relatively narrowfloodplain (410 ± 31 m), floodplain mining tailing deposits wereabsent, and dissolved copper concentrations were well belowacute criteria even during periods of high discharge (Fig. 3).

Relationships between dissolved copper and discharge devel-oped for each segment separately (Fig. 2) showed that copper ex-posure differed widely among the study segments. Predictions ofdaily copper concentration derived from copper-discharge regres-sions revealed that the mean number of days annually when dis-solved copper concentration exceeded the acute level (13.4 g·L−1)over the 3-year study averaged 42 days for Segment A (range 31–60 days), 12 days in Segment B (range 5–44 days), and 0 days inSegment C (Fig. 3). Copper exposure also varied seasonally amongsegments, with mean concentrations highest in spring and summerand lowest in fall and winter (Fig. 3), and declined with distancedownstream. During spring, mean daily copper concentration inSegment A was 9.6 g·L−1 (range 6.2–12.2 g·L−1), 7.9 g·L−1 in Seg-ment B (range 6.4–12.2 g·L−1), and 5.1 g·L−1 in Segment C (range3.8–5.8 g·L−1). Copper exposure declined during summer monthsbut remained elevated; the mean value was 9.5 g·L−1 in SegmentA (range 6.2–14.6 g·L−1), 6.4 g·L−1 in Segment B (range 5.1–8.1 g·L−1), and 4.2 g·L−1 in Segment C (range 3.6–5.1 g·L−1).During fall and winter, dissolved copper concentrations were lowand generally similar across all segments and did not exceed thechronic criteria of 8.9 g·L−1 in any segment (USEPA 2007); themean was 4.7 g·L−1 in Segment A (range 4.1–5.5 g·L−1), 4.3 g·L−1

in Segment B (range 4.1–4.6 g·L−1), and 3.1 g·L−1 in Segment C(range 2.9–3.2 g·L−1). In contrast with the main stem, dissolvedcopper concentrations in the three tributaries tested ranged from0.0 to 1.7 g·L−1.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Mayfield et al. 5

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

Water temperature varied little among segments. Mean maximumwater temperatures during the summer (15 June – 15 September) wereabout 1 °C warmer in Segment C than in Segments A and B; weeklymeans were 16.1 to 18.4 °C in Segment A, 16.4 to 18.6 °C in SegmentB, and 18.1 to 18.6 °C in Segment C. Maximum water temperaturesacross years were similar among all segments (22.2 to 22.9 °C,below the upper lethal temperature for brown trout of 24.7 °C;Elliott 1981), as were the mean number of days exceeding 20 °C(A = 16 days, B = 13 days, C = 21 days). However, these maximumwater temperatures exceeded the upper lethal temperature ofwestslope cutthroat trout (19.6 °C; Bear et al. 2011).

Survival and movementA total of 6389 relocations were used to estimate survival and

movement for the 183 radiotagged brown trout and 70 westslopecutthroat trout during the 138-week-long study period. The meannumber of relocations was 35 times per individual during weeklysurveys. Mortality was confirmed for 134 brown trout and 65 cut-throat trout during the study (Table 1). Survival time of both spe-cies ranged widely, from 1 to 138 weeks, with mean survival timesper tagging event varying from 21 to 60 weeks for brown trout andfrom 11 to 41 weeks for westslope cutthroat trout (Table 1). Fiftybrown trout (26.8%) and five westslope cutthroat trout (7.1%) werealive at the end of the study, surviving from 37 to 138 weekspost-tagging.

Seasonal survival patterns calculated from Kaplan–Meier survi-vorship curves were similar for both species across years, showinghighest mortality in the spring during elevated discharge (April–June), moderate mortality in summer (July–September), and littleto no mortality in fall and winter (Fig. 4). Annual survival of west-slope cutthroat trout from Kaplan–Meier analysis was low (0.09to 0.29), about half that of brown trout (0.38 to 0.61). Among-year survival patterns were similar for both species (i.e., sur-vival was relatively low in 2010, intermediate in 2009, and highin 2011).

The combined season spring and summer model yielded thebest-fitting model from the multistate model analysis, and AICcvalues of other models were >26.4, suggesting virtually no sup-port for other models (Table A1). Detection probability, an impor-tant consideration in multistate mark–recapture analyses (Perryet al. 2010), was high for both species with a weekly detectionprobability of 0.56 (0.01 SE) and a 0.96 (<0.01 SE) probability ofbeing detected at least once over 4 weeks. Additionally, the prob-ability that fish remained in the study area and were relocatedafter tagging was also very high (0.99, <0.01 SE).

Estimated survival rates calculated from the combined seasonmodel varied substantially by species, location, and season. Weeklymean survival rates across all locations were 0.981 for brown trout(range 0.970 to 0.992) and 0.927 for westslope cutthroat trout(range 0.758 to 0.996; Table 2). For both species, seasonal survival

rates were lowest during the spring–summer period (mean 0.550for brown trout, range 0.467 to 0.653; mean 0.199 for westslopecutthroat trout, range, 0.085 to 0.378) when dissolved copper con-centrations were elevated (>10 g·L−1; Fig. 3). Survival rates wereconsiderably higher in the fall–winter period when copper con-centrations were low (3–6 g·L−1), especially for westslope cut-throat trout (brown trout: mean survival 0.673, range 0.554 to0.814; westslope cutthroat trout: mean survival 0.438, range 0.001to 0.905).

Annual survival rates across all locations were low for bothspecies but especially so for westslope cutthroat trout (Table 2).Mean annual survival was 0.38 for brown trout (range 0.273 to0.531) and 0.12 for westslope cutthroat trout (range 0.001 to 0.342),about one-third that of brown trout. Annual survival rates fromthe multistate modeling and Kaplan–Meier estimations were sim-ilar. Lowest seasonal and annual survival for both species oc-curred in mainstem Segment A, which had the highest amountand percentage of floodplain mine tailings and frequent ex-ceedance of acute copper criteria during high discharges. In thissegment, mean annual survival was 0.273 for brown trout and0.001 for westslope cutthroat trout. In contrast, survival rates forboth species were considerably higher in Segment C, which hadno floodplain mining deposits and low dissolved copper concen-trations. The probability of survival in Segment C was 2.1 timeshigher than that in Segment A for brown trout (0.530 versus 0.273)and 122 times higher for cutthroat trout (0.122 versus 0.001).Brown trout survival in Segment B, with intermediate levels ofcopper contamination, was also intermediate to the rate in othermainstem segments, averaging 0.354. Highest westslope cut-throat trout survival was observed in Segment B (0.342). Survivalrates for westslope cutthroat trout in tributaries, where there waslittle or no dissolved copper, were nearly as low as in the mostcontaminated segment (A), especially during the spring–summerperiod (0.085). In contrast, brown trout survival in tributaries wasmuch higher, comparable to survival rates for Segment B (0.343).Mortality was confirmed for 29.6% of brown trout (8 of 27) and57.1% of cutthroat trout (16 of 28) that exhibited a distinct move-ment from the main stem into tributaries during respectivespawning migrations.

Estimates from the population models suggested that browntrout survived longer than westslope cutthroat trout in almostall segments and seasonal periods (brown trout range 18.7 to23.5 weeks; cutthroat trout range 2.9 to 26.0 weeks; Table 3).Brown trout occupying Segment A at the start of each seasonhad the lowest life expectancy (mean 18.7 weeks in the spring–summer period and 20.6 weeks in the fall–winter period), about2 weeks less than that in other segments. Westslope cutthroattrout life expectancy was lowest in Segment A and in tributar-ies (mean 9.1 weeks, range 2.9 to 15.5 weeks), less than half thatof other segments (mean 19.5 weeks, range 15.0 to 26.0 weeks).

Movement among segments was relatively rare and varied byspecies. For both species, fish had a high probability of remainingin the stream segment in which they were initially located at thebeginning of each season (Table 3). For brown trout, the probabilityof remaining in the same segment was similar across all segmentsand seasons, ranging from 0.936 to 0.980·week−1 (Table A2). West-slope cutthroat trout moved more frequently, and the weeklyprobability of residency in the same segment ranged from 0.653 to0.985. Movement to other segments was highest during spawning,and westslope cutthroat trout moved from the main stem totributaries during the spring–summer period (mean residencetime 4.0–4.6 weeks) and brown trout moved into tributariesduring the fall–winter period (mean residence time 1.6 to4.6 weeks).

Although both species usually remained in the original taggingsegment, model analysis suggested that both species typicallyspent some time in all other segments (Table 3). Brown trout fromeach segment were also found in three other locations (mean

Table 1. Number (percent) of survivors and mean survival time(weeks) of brown trout and cutthroat trout by radiotagging implanta-tion events in the upper Clark Fork River, Montana, 2009–2011.

YearNo.tagged

No.survived (%)

Mean survival timein weeks (range)

Brown trout 2009 (S) 70 19 (27.1) 60 (2–138)2010 (S) 59 12 (20.3) 34 (3–87)2010 (F) 45 13 (28.9) 37 (4–65)2011 (S) 9 5 (23.8) 21 (5–34)Total 183 49 (26.8) x = 38

Cutthroat trout 2009 (S) 21 1 (4.8) 31 (4–136)2010 (S) 33 2 (6.1) 18 (1–86)2010 (F) 5 2 (40.0) 41 (9–65)2011 (S) 11 0 (0.0) 11 (3–24)Total 70 5 (7.1) x = 25

Note: “S” refers to spring (April) tagging and “F” (September) to fall tagging.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

6 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

Fig. 4. Kaplan–Meier monthly survivorship curves with 95% confidence intervals for radiotagged brown trout and westslope cutthroat trout in the upper Clark Fork River, Montana, 2009–2011.

Pagination

notfinal(cite

DO

I)/

Pagination

provisoire(citer

leD

OI)

Mayfi

eldet

al.7

Publish

edby

NR

CR

esearchPress

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

1.3 weeks, range 0.7 to 2.5 weeks). There was no evidence foravoidance of Segment A by brown trout or of a greater tendency tomove during the spring–summer season when discharge and cop-per concentrations were elevated. However, westslope cutthroattrout from Segments B, C, and T rarely moved into Segment A(mean residence in Segment A: 0.4 weeks per season; range 0 to0.7 weeks).

DiscussionMining effects on water and habitat quality have the poten-

tial to affect fish survival and distribution over large areas, butpopulation-level effects from contamination exposure are noto-riously difficult to quantify (Luoma and Rainbow 2008; Hamiltonet al. 2016). Demographic responses to contaminants in the pasthave been estimated indirectly by relating in situ water contami-nation levels to laboratory-derived toxicity experiments, in situcage survival studies, tissue burdens from wild fish, and behav-ioral avoidance thresholds (Farag et al. 2003; Luoma et al. 2008). To

our knowledge, our study is the first to relate degree of contami-nant exposure to direct estimates of fish survival and movementat a population level over a large spatial scale.

Our results demonstrated that annual and seasonal survival forboth adult brown trout and cutthroat trout were strongly associ-ated with the distinct spatial and temporal differences in level ofcopper exposure among study segments. Annual survival ratesamong mainstem study segments of brown trout (0.27–0.53), andespecially those of westslope cutthroat trout (0.001–0.34), weregenerally well below expected. For example, Vincent (1987) re-ported an annual survival of 0.67 for adult brown trout in theunimpaired Madison River in southwestern Montana. In a reviewof trout survival rates in Rocky Mountain (USA) rivers, Carlson andRahel (2007) reported a mean annual survival rate of 0.43 (95% CI:0.38–0.48) for cutthroat trout and 0.44 (95% CI: 0.36–0.52) forbrown trout (see also Cook et al. 2015). Survival rates across studyyears were notably the lowest in the upstream segment (A) wheredissolved copper concentration and percentage of floodplain

Table 2. Weekly, seasonal, and annual estimates of survival (standard error (SE) in parentheses) ofbrown trout and westslope cutthroat trout in the three mainstem segments (A, B, C) and in tributaries(T) of the upper Clark Fork River, Montana, during two combined seasonal periods of spring andsummer (SS) and fall and winter (FW).

Segment Season Weekly Seasonal Annual

Brown trout A SS 0.971 (0.004) 0.467 (0.051)FW 0.980 (0.005) 0.584 (0.079) 0.273 (0.047)

B SS 0.975 (0.005) 0.519 (0.068)FW 0.985 (0.006) 0.681 (0.100) 0.354 (0.004)

C SS 0.984 (0.003) 0.653 (0.060)FW 0.992 (0.003) 0.814 (0.067) 0.531 (0.010)

T SS 0.978 (0.008) 0.560 (0.118)FW 0.981 (0.005) 0.613 (0.087) 0.343 (0.087)

Cutthroat trout A SS 0.912 (0.035) 0.090 (0.092)FW 0.758 (0.175) 0.001 (0.004) <0.001 (<0.001)

B SS 0.963 (0.015) 0.378 (0.151)FW 0.996 (0.002) 0.905 (0.040) 0.342 (0.139)

C SS 0.947 (0.011) 0.244 (0.074)FW 0.974 (0.012) 0.499 (0.164) 0.122 (0.055)

T SS 0.909 (0.019) 0.085 (0.047)FW 0.960 (0.038) 0.347 (0.355) 0.029 (0.034)

Table 3. Estimated mean number of weeks (SE in parentheses) brown trout and westslope cutthroattrout would occur in each of three mainstem segments (A, B, C) and in tributaries (T) of the upper ClarkFork River, Montana, during two 26-week periods of spring and summer (SS) and fall and winter (FW).

Mean weeks per segment

Segmentat start Season A B C T Total

Brown trout A SS 17.4 (0.8) 0.6 (0.1) <0.1 (<0.1) 0.7 (0.3) 18.7FW 16.2 (0.9) 0.6 (0.1) 0.9 (<0.1) 2.9 (0.3) 20.6

B SS 0.6 (0.4) 17.5 (1.2) 0.9 (0.5) 0.7 (0.5) 19.7FW 1.7 (0.2) 14.9 (0.2) 0.6 (<0.1) 4.6 (0.3) 21.8

C SS 0.4 (0.5) 0.9 (0.4) 19.2 (1.1) 1.0 (0.5) 21.5FW 0.8 (0.1) 0.4 (0.1) 20.7 (0.8) 1.6 (0.1) 23.5

T SS 1.1 (0.2) 2.1 (1.3) 0.1 (0.1) 18.7 (3.2) 22.0FW 2.8 (0.8) 2.2 (0.7) 2.5 (0.2) 13.8 (1.1) 21.3

Cutthroat trout A SS 5.3 (1.4) 0.4 (0.3) 2.3 (0.9) 4.6 (1.7) 12.6FW 5.4 (3.8) 0.0 0.0 0.0 5.4

B SS 0.7 (0.6) 8.8 (1.6) 2.1 (0.8) 4.6 (1.4) 16.2FW 0.0 26.0 (0.0) 0.0 0.0 26.0

C SS 0.3 (0.6) 0.6 (0.4) 10.1 (1.2) 4.0 (1.2) 15.0FW 0.0 0.0 19.4 (2.8) 0.0 19.4

T SS 0.2 (0.2) 0.9 (0.6) 5.0 (1.6) 9.4 (2.6) 15.5FW 0.2 (0.3) 0.7 (0.8) 0.0 2.0 (0.1) 2.9

Note: Bold values represent the time spent (weeks) in the same segment from the start to end of each 26-weektime period. Total represents the sum of weeks fish would likely survive across all segments.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

8 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

mine tailing deposits were greatest. In this 63 km-long upstreamsegment, there were over 63 ha of floodplain tailings, and dis-solved copper concentrations frequently exceeded the EPA acutecriteria threshold.

In contrast, annual survival rates were substantially higher forbrown trout and especially greater for cutthroat trout in down-stream river segments with little or no floodplain copper depositsand substantially reduced acute and chronic exposure to dis-solved copper. Catch-curve analysis of brown trout survival in theupper Clark Fork River in 2013–2015 documented a similar pat-tern of survival, and survival rates were lowest in Segment A (0.35)and increased with distance downstream (0.54, Segment B; 0.68,Segment C) (Cook et al. 2015).

The strong spatial gradient in dissolved copper is similar toresults from previous studies in the upper Clark Fork River show-ing an exponential downstream decline in copper accumulationsin sediment and in macroinvertebrate and fish tissues in relationto upstream floodplain source deposits (Axtmann and Luoma1991; Farag et al. 1995; Hornberger et al. 2009). Study results werealso consistent with prior evidence for mobilization of dissolved cop-per from floodplain deposits in this system during high discharges inspring and summer (Brooks and Moore 1989; Hornberger et al.2009). Particularly noteworthy was evidence for exceedance of theacute criteria threshold for dissolved copper, ranging from 31–60to 5–44 days in the upper and middle river segments, respectively,despite remediation efforts upstream.

Temporal differences in survival provided further support of alinkage between survival rates and copper exposure level. In all3 study years, Kaplan–Meier survival estimates for both specieswere lowest when peak pulses of copper exceeded acute levelsduring elevated spring discharge, and survival remained low dur-ing summer months when copper levels were chronically ele-vated. In contrast, survival was high in all segments during falland winter when copper levels were much lower (<5 g·L−1), evenin Segment A, the most contaminated upstream segment with thelowest spring–summer survival rates. Multistate model seasonalsurvival estimates exhibited a similar pattern, and mean fall–winter survival across all segments was 22% higher than spring–summer survival for brown trout and 54% higher for westslopecutthroat trout. Seasonal survival estimates generally increaseddownstream coincident with decreasing copper exposure, regard-less of estimator.

The observed seasonal survival patterns of upper Clark ForkRiver trout were unusual, as periods of high mortality in riverineadult trout populations generally occur during winter (Needhamet al. 1945; Carlson and Letcher 2003), immediately postspawning(DeRito et al. 2010), or during summer low flow (Elliott et al. 1997).In our study, fall–winter survival for both species was higher thanspring–summer survival, and we did not observe a pronouncedpostspawning mortality in brown trout in the fall during spawn-ing. Similarly, though some postspawning mortality of cutthroattrout was observed in late June after peak spawning (see below),the majority of spring–summer mortalities occurred in the spring(April–mid-June) prior to spawning.

High mortality in trout was not limited to periods of exposureto acute levels of dissolved copper, and it appears chronic expo-sure to elevated copper may have also contributed to low survivalin the upper Clark Fork River. For example, chronic exposure totrace metals, especially copper, can lead to reduced growth, reducedfeeding rates, and reduced swimming performance (Atchison et al.1987; Marr et al. 1995; Woodward et al. 1995a; Farag et al. 2000).Other physiological impairments caused by copper toxicity thatcan decrease trout survival include reduction in chemoreceptorfunction (Hansen et al. 1999b; McIntyre et al. 2008) and inhibitionof lateral line development (Linbo et al. 2006). The latter responsehas been documented at dissolved copper concentrations as lowas 2.0 g·L−1 (Sandahl et al. 2007), a concentration below thatobserved in all mainstem study segments. Survival rates and life

expectancy estimates were substantially lower for westslope cut-throat trout than for brown trout in all segments, a finding ob-served in previous studies that suggests Oncorhynchus species aremore sensitive to trace metal contamination than brown trout are(Hansen et al. 1999a).

The lack of strong behavioral avoidance of river sections byeither species during elevated copper levels was unexpected. Ma-trix analysis showed that tagged brown and cutthroat trout spentthe majority of time in the segment of initial tagging, even whentagged in segments with high copper exposure levels. Previouslaboratory studies using salmonids without an exposure historyto copper demonstrated avoidance of waters with dissolved cop-per concentrations as low as 12.0 g·L–1 (Woodward et al. 1995b;Goldstein et al. 1999; Hansen et al. 1999a; Svecevicius 2001), wellbelow peak copper levels in Segments A and B. However, we didnot observe large-scale movements into uncontaminated tributar-ies or to downriver segments during periods of peak copper levels.We did observe movement into tributaries, but movement coin-cided with spawning timing rather than as an apparent avoidanceto elevated copper levels. Chronic exposure to elevated copper hasbeen shown to induce olfactory impairment and reduce avoid-ance to copper and, in some cases, may also result in preferencefor elevated copper levels equivalent to rearing exposure levels(Svecevicius 2001; Hansen et al. 1999a; McIntyre et al. 2008). Thesefactors may explain the unexpected lack of avoidance of highcopper levels in our study.

The apparent linkage between survival rate and the degree ofcopper exposure that we observed for both brown trout and west-slope cutthroat trout in the upper Clark Fork River suggests thatfurther reduction of dissolved copper levels could improve sur-vival rates. Removal of tailings deposits in the uppermost 7 km ofSegment A in the early 1990s reduced accumulation of copper inwater, sediments, and macroinvertebrate tissues (Hornbergeret al. 2009). Our findings suggest that more extensive removal oftailings deposits in this segment would result in further decreasesin copper concentration both in this segment and farther down-stream (Luoma et al. 2008; Hornberger et al. 2009) and improvetrout survival rates accordingly.

The unexpectedly low survival rates of adult brown trout andwestslope cutthroat trout in uncontaminated tributaries, espe-cially during their respective spawning seasons, indicated othermortality sources beyond metal exposure are important in thesystem. Tributaries in the upper Clark Fork River have multipleanthropogenic factors that are likely affecting mortality, includ-ing unscreened irrigation canals, dewatering, and presence of mi-gration barriers at irrigation diversion dams (Saffel et al. 2011). Wefound that cutthroat trout have particularly high tributary mor-tality and are susceptible to entrainment in unscreened irrigationcanals in the late spring – early summer when returning to themain stem after spawning (Mayfield 2013). Improved fish passageand flows during spawning and rearing, therefore, could poten-tially increase survival and trout recruitment in the main stem,although the scope for recovery is unknown because we did notquantify these factors in our study.

There are several possible confounding factors that could haveinfluenced our results. First, we focused on the effects of copper asthe main factor affecting survival as it occurs in the most lethalconcentrations, but Clark Fork water and sediments contain acomplex mix of other trace metals (Marr et al. 1995; Hornbergeret al. 2009), and the interacting effects of multiple metals andother synergistic factors in the system (water temperature, pH)are not well understood (Cusimano et al. 1986; Cook et al. 2015).However, in situ survival studies of caged brown trout generallymirrored our findings, showing acute exposure to copper duringpeak flows and delayed mortality occurring during the declininglimb of the hydrograph in late June coincident with a rise in watertemperatures (Cook et al. 2015). Second, tagging could also haveaffected survival of fish already stressed from metal exposure.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Mayfield et al. 9

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

However, we did not observe unusually high mortality in themonth following tagging, and mean survival times post-taggingwere 23 and 44 months for westslope cutthroat trout and browntrout, respectively. Moreover, the similarity in the spatial patternof survival rates derived independently from catch-curve analysisfurther suggests that tagging did not significantly bias survivalestimates.

In sum, our results support the hypothesis that continued ex-posure to acute and chronically elevated dissolved copper concen-trations is a primary factor for the incomplete recovery of troutpopulations in the upper Clark Fork River. Moreover, our findingssuggest that survival could be substantially improved for browntrout and westslope cutthroat trout by additional removal of re-maining floodplain tailings. Future research on quantifying mor-tality factors in tributaries during spawning migrations and onidentifying sources of brown trout and westslope cutthroat troutrecruitment would aid remediation efforts for alleviating limitingfactors in the system.

AcknowledgementsPrimary funding for the project was provided by the Montana

Department of Justice, Natural Resource Damage Program, andMontana Fish, Wildlife and Parks. We thank Carol Fox, Tom Mo-stad, and Pat Cuneen of Natural Resource Damage Program fortheir strong financial and administrative support of the project.Additional support was provided by an EPA Science to AchieveResults graduate fellowship to Mariah Mayfield. Ben Whiteford,William Schreck, Ryan Kreiner, Brady Miller, Tracy Wendt, andColin Cooney were instrumental in collecting fish relocation datain often difficult conditions. We also thank Michael Colvin and JoeNaughton for technical assistance and Kurt Heim for his helpfulmanuscript review. The study was performed under the auspicesof the Montana State University Institutional Animal Care and UseCommittee Protocol 2009–2006. Any use of trade, firm, or productnames in the study is for descriptive purposes only and does notimply endorsement by the US Government.

ReferencesAnderson, R.O., and Neumann, R.M. 1996. Length, weight, and associated struc-

tural indices. In Fisheries techniques, second edition. Edited by B.R. Murphyand D.W. Willis. American Fisheries Society, Bethesda, Maryland.pp. 447–482.

Arnason, A.N. 1972. Parameter estimates from mark–recapture experiments ontwo populations subject to migration and death. Res. Pop. Ecol. 13: 97–113.doi:10.1007/BF02521971.

Arnason, A.N. 1973. The estimation of population size, migration rates, andsurvival in a stratified population. Res. Pop. Ecol. 15: 1–8. doi:10.1007/BF02510705.

Atchison, G.J., Henry, M.G., and Sandheinrich, M.B. 1987. Effects of metals onfish behavior: a review. Env. Biol. Fish. 18: 11–25. doi:10.1007/BF00002324.

Axtmann, E.V., and Luoma, S.N. 1991. Large-scale distribution of metal contam-ination in the fine-grained sediments of the Clark Fork River, Montana,U.S.A. Appl. Geochem. 6(1): 75–88. doi:10.1016/0883-2927(91)90064-V.

Axtmann, E.V., Cain, D.J., and Luoma, S.N. 1997. Effect of tributary inflows on thedistribution of trace metals in fine-grained bed sediments and benthic in-sects of the Clark Fork River, Montana. Env. Sci. Tech. 31: 750–758. doi:10.1021/es960417y.

Bear, E.A., McMahon, T.E., and Zale, A.V. 2011. Comparative thermal require-ments of westslope cutthroat trout and rainbow trout: implications for spe-cies interactions and development of thermal protection standards. Trans.Am. Fish. Soc. 136(4): 1113–1121. doi:10.1577/T06-072.1.

Brooks, R., and Moore, J.N. 1989. Sediment-water interactions in the metal-contaminated floodplain of the Clark Fork River, Montana, U.S.A. GeoJour-nal, 19: 27–36.

Brownie, C., Hines, J.E., Nichols, J.D., Pollock, K.H., and Hestbeck, J.B. 1993.Capture–recapture studies for multiple strata including non-Markovian tran-sitions. Biometrics, 49(4): 1173–1187. doi:10.2307/2532259.

Buchanan, R.A, and Skalski, J.R. 2010. Using multistate mark–recapture methodsto model adult salmonid migration in an industrialized river. Ecol. Model.221: 582–589. doi:10.1016/j.ecolmodel.2009.11.014.

Burnham, K.P., and Anderson, D.R. 2002. Model selection and multimodelinference: a practical information-theoretic approach. Springer-Verlag, NewYork.

Carlson, A.J., and Rahel, F.J. 2007. A basinwide perspective on entrainment of

fish in irrigation canals. Trans. Am. Fish. Soc. 136(5): 1335–1343. doi:10.1577/T06-111.1.

Carlson, S.M., and Letcher, B.H. 2003. Variation in brook and brown trout sur-vival within and among seasons, species, and age classes. J. Fish Biol. 63(3):780–794. doi:10.1046/j.1095-8649.2003.00191.x.

Caswell, H. 2001. Matrix population models: construction, analysis and interpre-tation. 2nd edition. Sinauer Associates, Sunderland, Massachusetts.

Clark Fork Coalition. 2011. Aquatic restoration strategy for the upper Clark ForkBasin: elements of an integrated approach [online]. Available from https://clarkfork.org/wp-content/uploads/2014/07/CFC-UCF_AquaticRestStrat_2011.pdf.

Clements, W.H. 2000. Integrating effects of contaminants across levels of bio-logical organization: an overview. J. Aquat. Ecosys. Stress Recov. 7(2): 113–116.doi:10.1023/A:1009927612391.

Cook, N., Saffel, P., Liermann, B., Lindstrom, J., and Selch, T. 2015. Upper ClarkFork River fisheries monitoring study: 2014 annual report. Montana Fish,Wildlife and Parks report, Missoula, Montana.

Cusimano, R.F., Brakke, D.F., and Chapman, G.A. 1986. Effects of pH on thetoxicities of cadmium, copper, and zinc to steelhead trout (Salmo gairdneri).Can. J. Fish. Aquat. Sci. 43(8): 1497–1503. doi:10.1139/f86-187.

DeRito, J.N., Zale, A.V., and Shepard, B.B. 2010. Temporal reproductive separa-tion of fluvial Yellowstone Cutthroat Trout from Rainbow Trout and hybridsin the Yellowstone River. N. Am. J. Fish. Manage. 30(4): 866–886. doi:10.1577/M09-006.1.

Elliott, J.M. 1981. Some aspects of thermal stress on freshwater teleosts. In Stressand fish. Edited by A.D. Pickering. Academic Press, New York. pp. 209–245.

Elliott, J.M., Hurley, M.A., and Elliott, J.A. 1997. Variable effects of droughts onthe density of sea-trout Salmo trutta populations over 30 years. J. Appl. Ecol.34(5): 1229–1238. doi:10.2307/2405234.

Esri. 2009. ArcMap 9.3.1. Environmental Systems Resource Institute, Redlands,California.

Farag, A.M., Stansbury, M.A., Hogstrand, C., MacConnell, E., and Bergman, H.L.1995. The physiological impairment of free-ranging brown trout exposed tometals in the Clark Fork River, Montana. Can. J. Fish. Aquat. Sci. 52: 2023–2050.

Farag, A.M., Suedkamp, M.J., Meyer, J.S., Barrows, R., and Woodward, D.F. 2000.Distribution of metals during digestion by cutthroat trout fed benthic inver-tebrates contaminated in the Clark Fork River, Montana and the Coeurd’Alene River, Idaho, U.S.A., and fed artificially contaminated Artemia. J. FishBiol. 56: 173–190. doi:10.1111/j.1095-8649.2000.tb02093.x.

Farag, A.M., Skaar, D., Nimick, D.A., MacConnell, E., and Hogstrand, C. 2003. Char-acterizing aquatic health using salmonid mortality, physiology, and biomassestimates in streams with elevated concentrations of arsenic, cadmium, copper,lead, and zinc in the Boulder River watershed, Montana. Trans. Am. Fish. Soc.132: 450–467. doi:10.1577/1548-8659(2003)132<0450:CAHUSM>2.0.CO;2.

Goldstein, J.N., Woodward, D.F., and Farag, A.M. 1999. Movements of adult Chi-nook salmon during spawning migration in a metals-contaminated system,Coeur d’Alene River, Idaho. Trans. Am. Fish. Soc. 128(1): 121–129. doi:10.1577/1548-8659(1999)128<0121:MOACSD>2.0.CO;2.

Hamilton, P.B., Cowx, I.G., Oleksiak, M.F., Griffiths, A.M., Grahn, M., Stevens, J.R.et al. 2016. Population-level consequences for wild fish exposed to sublethalconcentrations of chemicals — a critical review. Fish Fisher. 17(3): 545–566.doi:10.1111/faf.12125.

Hansen, J.A., Woodward, D.F., Little, E.E., DeLonay, A.J., and Bergman, H.L.1999a. Behavioral avoidance: possible mechanisms for explaining abundanceand distribution of trout species in a metal-impacted river. Env. Toxic. Chem.18(2): 313–317. doi:10.1002/etc.5620180231.

Hansen, J.A., Rose, J.D., Jenkins, R.A., Gerow, K.G., and Bergman, H.L. 1999b.Chinook salmon (Oncorhynchus tshawytscha) and rainbow trout (Oncorhynchusmykiss) exposed to copper: neurophysiological and histological effects onthe olfactory system. Env. Toxic. Chem. 18(9): 1979–1991. doi:10.1002/etc.5620180917.

Hestbeck, J.B., Nichols, J.D., and Malecki, R.A. 1991. Estimates of movement andsite fidelity using mark–resight data of wintering Canada geese. Ecology 72:523–533.

Hornberger, M.I., Luoma, S.N., Johnson, M.L., and Holyoak, M. 2009. Influence ofremediation in a mine-impacted river: metal trends over large spatial andtemporal scales. Ecol. Applic. 19(6): 1522–1535. doi:10.1890/08-1529.1.

Jackson, C.H. 2011. Multi-state models for panel data: The msm package for R. J.Stat. Soft. 38(8): 1–29. doi:10.18637/jss.v038.i08.

Kaplan, E.L., and Meier, P. 1958. Nonparametric estimation from incompleteobservations. J. Am. Stat. Assoc. 53: 457–481. doi:10.1080/01621459.1958.10501452.

Kiser, T., Hansen, J., and Kennedy, B. 2010. Impacts and pathways of mine con-taminants to bull trout (Salvelinus confluentus) in an Idaho watershed. Arch.Env. Contam. Toxic. 59: 301–311. doi:10.1007/s00244-009-9457-x.

Linbo, T.L., Stehr, C.M., Incardona, J.P., and Scholz, N.L. 2006. Dissolvedcopper triggers cell death in the peripheral mechanosensory system oflarval fish. Environ. Toxicol. Chem. 25(2): 597–603. doi:10.1897/05-241R.1.PMID:16519324.

Lindstrom, J. 2011. Upper Clark Fork River fish sampling: 2008–2010. MontanaFish, Wildlife and Parks, Helena, Montana.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

10 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

Luoma, S.N., and Rainbow, P.S. 2008. Metal contamination in aquaticenvironments: science and lateral management. Cambridge University Press,New York.

Luoma, S.N., Moore, J.N., Farag, A.M., Hillman, T.H., Cain, D.J., and Hornberger,M. 2008. Mining impacts on fish in the Clark Fork River, Montana: a fieldecotoxicology case study. In The toxicology of fishes. Edited by R.T. DiGuilioand D.E. Hinton. CRC Press, Boca Raton, Florida. pp. 779–804.

Marr, J.C.A., Bergman, H.L., Parker, M., Lipton, J., Cacela, D., Erickson, W., andPhillips, G.R. 1995. Relative sensitivity of brown and rainbow trout to pulsedexposures of an acutely lethal mixture of metals typical of the Clark ForkRiver, Montana. Can. J. Fish. Aquat. Sci. 52(9): 2005–2015. doi:10.1139/f95-792.

Mayfield, M.P. 2013. Limiting factors for trout populations in the upper ClarkFork River Superfund site, Montana. M.Sc. thesis, Montana State University,Bozeman, Montana.

McIntyre, J.K., Baldwin, D.H., Meador, J.P., and Scholz, N.L. 2008. Chemosensorydeprivation in juvenile coho salmon exposed to dissolved copper under vary-ing water chemistry conditions. Env. Sci. Tech. 42: 1352–1358. doi:10.1021/es071603e.

Moore, J.N., and Luoma, S.N. 1990. Hazardous wastes from large-scale metalextraction. Environ. Sci. Technol. 24: 1279–1285.

Needham, P.R., Moffett, J.W., and Slater, D.W. 1945. Fluctuations in wild browntrust populations in Convict Creek, California. J. Wild. Manage. 9(1): 9–25.doi:10.2307/3795940.

Nimick, D.A., and Moore, J.N. 1991. Prediction of water-soluble metal concentra-tions in fluvially deposited tailings sediments, Upper Clark Fork Valley, Mon-tana, U.S.A. Appl. Geochem. 6: 635–646. doi:10.1016/0883-2927(91)90074-Y.

Perry, R.W, Skalski, J.R., Brandes, P.L., Sandstrom, P.T., Klimley, A.P.,Ammann, A., and MacFarlane, B. 2010. Estimating survival and migrationroute probabilities of juvenile Chinook salmon in the Sacramento–San JoaquinRiver Delta. N. Am. J. Fish. Manage. 30: 142–156. doi:10.1577/M08-200.1.

Phillips, G., and Lipton, J. 1995. Injury to aquatic resources caused by metals inMontana’s Clark Fork River basin: historic perspective and overview. Can. J.Fish. Aquat. Sci. 52: 1990–1993. doi:10.1139/f95-190.

Powell, L.A. 2007. Approximating variance of demographic parameters usingthe delta method: a reference for avian biologists. Condor, 109(4): 949–954.doi:10.1650/0010-5422(2007)109[949:AVODPU]2.0.CO;2.

R Development Core Team. 2010. R: a language and environment for statisticalcomputing [online]. R Foundation for Statistical Computing, Vienna, Austria.Available from http://www.r-project.org.

Ross, M.J., and Kleiner, C.F. 1982. Shielded-needle technique for surgically im-

planting radio-frequency transmitters in fish. Prog. Fish-Cult. 44: 41–43. doi:10.1577/1548-8659(1982)44[41:STFSIR]2.0.CO;2.

Saffel, P., Liermann, B., Lindstrom, J., Knotek, L., Mostad, T., and Fox, C. 2011.Prioritization of areas in the Upper Clark Fork River basin for fishery en-hancement [online]. Montana Fish, Wildlife, and Parks report. Missoula,Montana. Available from https://dojmt.gov/lands/upper-clark-fork-river-basin/.

Sandahl, J.F., Baldwin, D.H., Jenkins, J.J., and Scholz, N.L. 2007. A sensory systemat the interface between urban stormwater runoff and salmon survival. Env.Sci. Tech. 41(8): 2998–3004. doi:10.1021/es062287r.

Stubben, C.J., and Milligan, B.G. 2007. Estimating and analyzing demographicmodels using the popbio package in R. J. Stat. Soft. 22: 1–23. doi:10.18637/jss.v022.i11.

Svecevicius, G. 2001. Avoidance response of rainbow trout Oncorhynchus mykiss toheavy metal mixtures: a comparison with acute toxicity tests. Bull. Env.Contam. Toxic. 67: 680–687. doi:10.1007/s001280177.

USEPA. 2007. Aquatic life ambient freshwater quality criteria — copper. USEnvironmental Protection Agency Report 822-R-07-001.

Vincent, E.R. 1987. Effects of stocking catchable-size hatchery rainbow trout ontwo wild trout species in the Madison River and O’Dell Creek, Montana. N.Am. J. Fish. Manage. 7: 91–105. doi:10.1577/1548-8659(1987)7<91:EOSCHR>2.0.CO;2.

Weigel, D.E., Peterson, J.T., and Spruell, P. 2002. A model using phenotypiccharacteristics to detect introgressive hybridization in wild westslope cut-throat trout and rainbow trout. Trans. Am. Fish. Soc. 131: 389–403. doi:10.1577/1548-8659(2002)131<0389:AMUPCT>2.0.CO;2.

White, G.C., and Burnham, K.P. 1999. Program MARK: survival estimation frompopulations of marked animals. Bird Study, 46(Suppl.): 120–138. doi:10.1080/00063659909477239.

Woodward, D.F., Farag, A.M., Bergman, H.L., DeLonay, A.J., Little, E.E.,Smiths, C.E., and Barrows, F.T. 1995a. Metals-contaminated benthic inverte-brates in the Clark Fork River Montana: effects on age-0 brown trout andrainbow trout. Can. J. Fish. Aquat. Sci. 52(9): 1994–2004. doi:10.1139/f95-791.

Woodward, D.F., Hansen, J.A., Bergman, H.L., DeLonay, A.J., and Little, E.E.1995b. Brown trout avoidance of metals in water characteristic of the ClarkFork River, Montana. Can. J. Fish. Aquat. Sci. 52: 2031–2037. doi:10.1139/f95-794.

Zale, A.V., Brooke, C., and Fraser, W.C. 2005. Effects of surgically implantedtransmitter weights on growth and swimming stamina of small adultwestslope cutthroat trout. Trans. Am. Fish. Soc. 134: 653–660. doi:10.1577/T04-050.1.

Appendix A

Table A1. Model selection results for the 10 multistate models tested using various combinations of survival and movement rates.

Model

Movement rates Survival rates

AICc AICc AICc weightSpatial Temporal Spatial Temporal

1 Variable Spring–summer constant;fall–winter constant

Variable Spring–summer constant;fall–winter constant

14 489.41 0.00 1.00

2 Variable Spring separate; otherseasons constant

Variable Spring separate; otherseasons constant

14 515.80 26.39 0.00

3 Variable Constant Variable Constant 14 556.89 67.48 0.004 Constant Constant Variable Constant 14 566.89 77.49 0.005 Variable Variable seasons Variable Variable seasons 14 575.41 86.00 0.006 Variable Constant Variable Variable seasons 14 579.13 89.72 0.007 Constant Constant Variable Constant 14 634.61 145.21 0.008 Constant Constant Constant Constant 14 639.66 150.26 0.009 Constant Variable seasons Constant Variable seasons Did not converge10 Variable Variable seasons Constant Variable seasons Did not converge

Note: “Constant” indicates that the model held transition rate estimates equal for all locations segments (spatial) or seasonal time periods (temporal). “Variable”indicates models that allowed transition rate estimates to vary. “Did not converge” refers to models where maximum likelihood estimates could not be calculated,even using the alternative optimization method in Program MARK (White and Burnham 1999). For all models, survival and movement rates were estimatedindependently for each species.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Mayfield et al. 11

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.

Table A2. Estimated probability of weekly residency (SE in parentheses) of brown trout and cutthroat trout in thethree mainstem segments (A, B, C) and in tributaries (T) of the upper Clark Fork River, Montana, during two combinedseasonal periods of spring and summer (SS) and fall and winter (FW).

Segment at end

Segmentat start Season A B C T

Brown trout A SS 0.965 (0.004) 0.003 (0.001) 0.0 0.003 (0.001)FW 0.956 (0.008) 0.002 (0.002) 0.003 (0.005) 0.019 (0.005)

B SS 0.002 (0.002) 0.967 (0.006) 0.004 (0.006) 0.003 (0.002)FW 0.007 (0.004) 0.946 (0.011) 0.0 0.032 (0.009)

C SS 0.001 (0.001) 0.004 (0.002) 0.975 (0.004) 0.004 (0.002)FW 0.003 (0.002) 0.001 (0.001) 0.980 (0.005) 0.008 (0.004)

T SS 0.005 (0.005) 0.009 (0.006) 0.0 0.964 (0.010)FW 0.017 (0.005) 0.015 (0.005) 0.014 (0.005) 0.936 (0.010)

Cutthroat trout A SS 0.912 (0.051) 0.0 0.0 0.105 (0.042)FW 0.653 (0.175) 0.0 0.0 0.0

B SS 0.011 (0.010) 0.901 (0.023) 0.0 0.062 (0.020)FW 0.0 0.985 (0.002) 0.0 0.0

C SS 0.003 (0.003) 0.003 (0.003) 0.884 (0.004) 0.004 (0.002)FW 0.0 0.0 0.974 (0.012) 0.0

T SS 0.0 0.009 (0.006) 0.073 (0.018) 0.828 (0.026)FW 0.031 (0.031) 0.019 (0.019) 0.0 0.897 (0.041)

Note: Bold values are the probabilities of a fish remaining in the same segment at the start and end of the time period.

Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

12 Can. J. Fish. Aquat. Sci. Vol. 00, 0000

Published by NRC Research Press

Can

. J. F

ish.

Aqu

at. S

ci. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y T

hom

as M

cMah

on o

n 07

/19/

19Fo

r pe

rson

al u

se o

nly.